LIBERTAS MATHEMATICA, vol XXX (2010)

Green Function of the Laplacian for the

Neumann Problem in Rn+
E. CONSTANTIN and N. H. PAVEL
Abstract. By analogy with the Green function for the Dirichlet problem for a
domain Ω, one considers the Green type function for the Neumann problem for
Ω. We obtain the Green type function for the positive half-space of Rn and use
it to solve the Neumann problem. The cases where Ω is the positive half space
of R2 and of R3 are analyzed in details under new hypotheses.
Keywords: Green’s function for the Neumann problem, Neumann function,
Dirichlet problem.
Mathematics Subject Classification (2000): Primary 35J05; Secondary
35J25, 35J08.

1 Introduction
Let Ω be a domain of Rn with smooth boundary ∂Ω.
Recall the Dirichlet and Neumann problems for the Laplacian.
(1) Dirichlet Problem (DP):
Find u ∈ C 2 (Ω) ∩ C(Ω̄)) such that

∆u = f in Ω, u = g on ∂Ω. (1)

(2) Neumann Problem (NP):

Find u ∈ C 2 (Ω) ∩ C 1 (Ω̄) such that
du
∆u = f in Ω, = g1 on ∂Ω, (2)
dηx
du
where ηx is the outward unit normal to ∂Ω at x ∈ ∂Ω, is the normal derivative of u at
dηx
x, f : Ω → R, and g, g1 : ∂Ω → R are prescribed continuous functions.
By analogy with the Green function for the Dirichlet problem for a domain Ω, we consider
the Green type function for the Neumann problem for Ω (also known as Neumann’s function,
or Green’s function for the Neumann problem or Green’s function of the second kind). We
will give the explicit forms of the Neumann’s functions and the solution of the Neumann
problem for the upper half-space of the n-th dimensional euclidian space Rn , n ≥ 2.

57
58 E. CONSTANTIN AND N. H. PAVEL

The concept of Green type function for (NP) has been considered by several authors
([14], [9], [19], [10], [17], [15], [2], [6], [12], [13]).
In [3], there are presented the expressions of the Green’s functions for the Neumann’s
problem for a ball in R2 and R3 . In [12], Neumann’s function for the sphere in R3 is
constructed using the classical method of images and expressed in terms of eigenvalues
associated with the surface, leading to an analogue of the Poisson integral as a solution to
the Neumann problem for the sphere. In [6], there are given the Neumann’s function and
the solution of the Neumann problem for the interior and the exterior of the sphere of Rn ,
n ≥ 2.
Green functions of the Laplacian for Neumann problems relative to all domains bounded
by the coordinate surfaces in the circular cylindrical coordinate system are constructed in
[2].
In [13] it can be found a construction of the Green function for the three-dimensional
Laplace equation, in the interior of an arbitrary rectangular channel subject to homogeneous
Neumann conditions on the boundaries.
The explicit forms of the Green’s function and of the Neumann’s function G and G1 in
the half-plane x2 ≥ 0 in R2 are given in [10], while in [17] there are given the Green’s and
Neumann’s functions for both the half-plane x2 ≥ 0 and the upper half-space x3 ≥ 0 in R3 .
However, in both books it is not proved that the functions u and u1 obtained by means of
G and G1 and Green’s formula are solutions of the (DP) and (NP), respectively.
In [9], the explicit expressions of the Green’s and the Neumann’s functions are given for
the upper half-spaces R2+ and R3+ , and it is shown that the functions u and u1 obtained with
their aid are solutions of the (DP) and (NP), respectively, under the hypothesis that g and
g1 are analytical. Physical interpretations of the Green’s and Neumann’s functions can be
found in [9]. Also the Green’s and Neumann’s functions for the interior and the exterior of
the unit circle in R2 and unit sphere in R3 centered at the origin are given in [9].
In [19], it is constructed the Green’s function for the Neumann problem formulated for
the unit sphere in R3 centered at the origin. Also in [19], there are obtained the explicit
solutions u and u1 of the (DP) and (NP) using Green’s formula for the tridimensional
upper-half space, respectively (without using the Green’s function for (DP) or the Green’s
function for (NP)). Then it is proved that u and u1 are indeed solutions of (DP) and (NP),
respectively, under the assumptions: g and g1 are continuous on x3 = 0, g is bounded and
|g(x1 , x2 )| ≤ √ 2 M 2 1+a , where 0 < a < 1, and M is a positive constant.
( x1 +x2 )

In [4], it is derived the Green’s function and it is shown a Poisson Formula for the positive
half-space of Rn , n ≥ 2.
We obtain the Green type function for the positive half-space Rn+ , n ≥ 2, and use it
to find the solution of the (NP) under the assumption that g is continuous and bounded
on Rn−1 , and g1 is continuous and with compact support in Rn−1 (subsection 3.3). The
cases where Ω is the positive half space R2+ and R3+ , respectively, are presented in details
under hypotheses different from the ones considered by other authors (subsection 3.1 and
subsection 3.2).
GREEN FUNCTION OF THE LAPLACIAN FOR THE NEUMANN PROBLEM 59

2 Green Functions for Dirichlet and Neumann Prob-

lems
Suppose that the function u ∈ C 2 (Ω)∩C 1 (Ω̄). Then the following (Riemann-Green) formula
holds, i.e., Z Z
du dG(x, y)
u(y) = G(x, y)∆u(x) dx − (G(x, y) (x) − u(x) ) dσx , (3)
Ω ∂Ω dη x dηx
where G(x, y) = ψ(kx − yk) + g̃(x), x ∈ Ω̄, y ∈ Ω, x 6= y, g̃(x) is an arbitrary harmonic
function in Ω and
r2−n 1
ψ(r) = , if n > 2 and ψ(r) = ln r, if n = 2, (3′ )
(2 − n)σn 2π
with r = kx − yk. Finally, σn is the surface area of the unit sphere in Rn .
Suppose that the function G(x, y) in the above formula satisfies the additional property
G(x, y) = 0 for x ∈ ∂Ω, y ∈ Ω. (4)
Then the solution u of (DP) with the regularity u ∈ C 2 (Ω) ∩ C 1 (Ω̄) (if any) is given by
Z Z
dG(x, y)
u(y) = G(x, y)f (x)dx + g(x) dσx . (5)
Ω ∂Ω dηx
A Green’s function G(x, y) for ∆ on Ω is a function G as above, i.e., having the properties
x → G(x, y) belongs to C 2 (Ω̄ \ {y}), ∆x G(x, y) = 0 for x ∈ Ω, G(x, y) = 0, for x ∈ ∂Ω,
y ∈ Ω.
For the Neumann problem, u is not prescribed on the boundary ∂Ω of Ω, so the formula
(3) suggests to look for a function G = G1 with the condition
dG1 (x, y)
= 0, for x ∈ ∂Ω, (6)
dηx
instead of G(x, y) = 0, for x ∈ ∂Ω for the Dirichlet problem. This means to find a function
g̃(x) = K(x, y) with x → K(x, y) in C 1 (Ω̄) ∩ C 2 (Ω), ∀ y ∈ Ω,

∆x K(x, y) = 0, ∀ x, y ∈ Ω,

dK(x, y) dψkx − yk x−y

=− = −ψ ′ (r) · ηx , x ∈ ∂Ω, y ∈ Ω,
dηx dηx r

r1−n
where ψ is given above, i.e., ψ ′ (r) = . Therefore, the solution u1 ∈ C 2 (Ω) ∩ C 1 (Ω̄) (if
σn
any) of the (NP) is necessarily given by
Z Z
u1 (y) = G1 (x, y)f (x) dx − G1 (x, y)g1 (x)dσx . (7)
Ω ∂Ω

Such a function G1 will be called a Green type function for the (NP) on Ω (also called
Green’s function for (NP) in [14], [19], [3], [2], or Neumann’s function in [9], [17], [15], or
Green’s function of the second kind in [10]).
60 E. CONSTANTIN AND N. H. PAVEL

3 Green function for the Neumann Problem for Rn+

In this section we will build Green’s functions for the (DP) and (NP) for the half-space Rn+ ,
n ≥ 2, using the ideas developed in Section 2. After that, we will check directly that the
corresponding representation formulas for the solutions of (DP) and (NP) are valid under
appropriate assumptions on g and g1 .
First we consider the cases n = 2 and n = 3 under hypothesis on g1 different from the
ones used by other authors.

3.1 Construction of the Green function for the Neumann Problem

for R2+
[14] Let Ω = {x = (x1 , x2 ) ∈ R2 , x2 > 0} be the positive half-space. Clearly, ∂Ω = {x =
(x1 , x2 ) ∈ R2 , x2 = 0} = {x = (x1 , 0), x1 ∈ R}. For y = (y1 , y2 ) define by reflection
y ∗ = (y1 , −y2 ). Then the function:

1
G(x, y) = (lnkx − yk − lnkx − y ∗ k), x ∈ Ω̄, y ∈ Ω, x 6= y, (8)
2π

is a Green function for the (DP), and

1
G1 (x, y) = (lnkx − yk + lnkx − y ∗ k), x ∈ Ω̄, y ∈ Ω, x 6= y, (9)
2π

is a Green type function for (NP).

As kx − yk = kx − y ∗ k = r for y ∈ Ω, x ∈ ∂Ω, it follows that G(x, y) = 0, x ∈ ∂ Ω, y ∈ Ω.
The outward normal ηx to ∂Ω = {x = (x1 , x2 ) ∈ R2 , x2 = 0} at x ∈ ∂Ω is ηx = (0, −1).
The normal derivative of G is:

dG ∂G
=− = π −1 y2 r−2 with x = (x1 , 0), y = (y1 , y2 ), y2 > 0,
dηx ∂x2
dG1
where r2 = (x1 − y1 )2 + y22 . Similarly we can check that = 0.
dηx
The formula (3) suggests that a solution to the problem: ∆u = 0 in Ω, u = g on ∂Ω,
could be
Z ∞
y2 g(x1 )dx1
Z
dG
u(y) = u(x) (x, y) dσx = π −1 2 2, (10)
∂Ω dηx −∞ (x1 − y1 ) + y2

with x = (x1 , 0), g(x1 , 0) = g(x1 ), y = (y1 , y2 ), y2 > 0.

1
Clearly G1 (x, y) = lnkx − yk, for x = (x1 , 0), y ∈ Ω. Therefore, (3) suggests that a
π
possible solution to the Neumann problem (with f = 0) could be
Z Z ∞
u1 (y) = − G1 (x, y)g1 (x)dσx = −π −1 g1 (x1 )lnkx − ykdx1 , (11)
∂Ω −∞
GREEN FUNCTION OF THE LAPLACIAN FOR THE NEUMANN PROBLEM 61

with x = (x1 , 0), g1 (x1 , 0) = g1 (x1 ), y = (y1 , y2 ), y2 > 0.

The functions g and g1 must guarantee the convergence of the improper integrals (10)
and (11), respectively (i.e., the existence of u and u1 ), and the fact that these functions u and
u1 are solutions of the above Dirichlet and Neumann problems with f = 0. An important
case in which these requirements are fulfilled is given by:

Theorem 1 Let g be continuous and bounded on R and g1 be continuous with compact

support in R. Then the functions u and u1 given by the improper integrals (10) and (11)
satisfy:
1. u ∈ C 2 (Ω) ∩ C(Ω̄)), ∆u = 0 in Ω, u = g on ∂Ω.
du1
2. u1 ∈ C 2 (Ω) ∩ C 1 (Ω̄), ∆u1 = 0 in Ω, = g1 , on ∂Ω.
dηx
Proof. The key fact is the elementary formula:
Z ∞
−1 y2 dx1
π = 1, x = (x1 , 0), y = (y1 , y2 ), y2 > 0. (12)
−∞ (x1 − y1 )2 + y22

which follows by the change of variable: x1 − y1 = y2 z, so dx1 = y2 dz.

The fact that u and u1 are harmonic in Ω follows by differentiating under the integral
sign, in conjunction with:
dG d
∆y (x, y) = (∆y G(x, y)), G(x, y) = G(y, x) for x ∈ Ω̄, y ∈ Ω, x 6= y, and
dηx dηx
∆y G(x, y) = ∆y G(y, x) = 0 for x ∈ ∂Ω, y ∈ Ω.
As g is bounded and g1 is continuous and with compact support in R, u and u1 given
by (10) and (11) are bounded.
Let us to prove that u(x) = g(x) for x = (x1 , 0), i.e., that

lim u(y) = g(y0 ), ∀ y0 with y0 = (y01 , 0), y = (y1 , y2 ), with y2 > 0. (13)
y→y0

In view of (10), we can write

∞
y2 (g(x1 ) − g(y01 ))dx1
Z
u(y) − g(y0 ) = π −1 , (14)
−∞ (x1 − y1 )2 + y22

y = (y1 , y2 ), y2 > 0.
Given ǫ > 0, there exists δ(ǫ) > 0 such that

|g(x1 ) − g(y01 )| < ǫ, for |x1 − y01 | < δ.

Let us write:
y2 (g(x1 ) − g(y01 ))dx1
Z
u(y) − g(y0 ) = π −1 +
|x1 −y10 |<δ (x1 − y1 )2 + y22

y2 (g(x1 ) − g(y01 ))dx1

Z
+π −1 = I1 + I2 . (15)
|x1 −y10 |≥δ (x1 − y1 )2 + y22
62 E. CONSTANTIN AND N. H. PAVEL

In view of (10) and (12), it follows that |I1 | ≤ ǫ. Moreover, by the same change of
variable as above, we have:
0 −δ−y
y1
Z y10 −δ Z 1
y2 dx1 y2 dz
= → 0, as y1 → y01 , y2 ↓ 0.
−∞ (x1 − y1 )2 + y22 −∞ 1 + z2

Similarly,
Z ∞ Z ∞
y2 dx1 dz
= → 0, as y1 → y01 , y2 ↓ 0.
y10 +δ (x1 − y1 )2 + y22 0 +δ−y
y1
y2
1 1 + z2

It follows that limy→y0 I2 = 0, as g is bounded on R. This in conjunction with |I1 | ≤ ǫ

implies lim supy→y0 |u(y) − g(y0 ) ≤ ǫ, ∀ǫ > 0, so (13) is valid.
Under the hypotheses on g1 , the function u1 is well defined and of class C 1 (Ω̄). Differ-
entiating under the integral sign we obviously obtain:
Z ∞
∂u1 (y) y2 g1 (x1 )dx1
− = π −1 2 2.
∂y2 −∞ (x 1 − y1 ) + y2

du1 ∂u1 (y)

On the basis of the previous discussion, = − this implies
= g1 (y) on the
dηy ∂y2
boundary ∂Ω of Ω as g1 is bounded, which completes the proof.

3.2 Construction of the Green function for the Neumann Problem

for R3+
We build the Green functions G and G1 for the (DP) and (NP), respectively, in the situation
where Ω = R3+ , and then we verify directly that the representation formulas of the functions
u and u1 obtained by means of G and G1 do indeed provide us with solutions of the (DP)
and (NP), respectively.
The following classical result will be needed to achieve that (Theorem 1.1.11, [1]).

Theorem 2 [1] Let f : Rn → (−∞, ∞) be a measurable function and let R > 0.

−λ n
Z ≤ ckxk , for all x ∈ R with kxk ≤ R, where c is a positive constant, and
i) If |f (x)|
λ < n, then |f (x)|dx < +∞.
kxk≤R
−λ n
ii) If |f (x)|
Z ≤ ckxk , for all x ∈ R with kxk ≥ R, where c is a positive constant, and
λ > n, then |f (x)|dx < +∞.
kxk≥R

Next we construct the Green function for the Neumann problem for the positive half-
space R3+ .
Let Ω = {x = (x1 , x2 , x3 ) ∈ R3 , x3 > 0} be the positive half-space. Clearly, ∂Ω = {x =
(x1 , x2 , x3 ) ∈ R3 , x3 = 0} = {x = (x1 , x2 , 0), (x1 , x2 ) ∈ R2 }. For y = (y1 , y2 , y3 ) define by
reflection y ∗ = (y1 , y2 , −y3 ). Then the function:
GREEN FUNCTION OF THE LAPLACIAN FOR THE NEUMANN PROBLEM 63

1 1
G(x, y) = − , x ∈ Ω̄, y ∈ Ω, x 6= y, (16)
4πkx − yk 4πkx − y ∗ k

is a Green function for the (DP) with f = 0, and

1 1
G1 (x, y) = + , x ∈ Ω̄, y ∈ Ω, x 6= y, (17)
4πkx − yk 4πkx − y ∗ k

is a Green type function for (NP) with f = 0.

As kx − yk = kx − y ∗ k = r for y ∈ Ω, x ∈ ∂Ω, it follows that G(x, y) = 0, x ∈ ∂Ω, y ∈ Ω.
The outward normal ηx to ∂Ω = {x = (x1 , x2 , x3 ) ∈ R3 , x3 = 0} at x ∈ ∂Ω is ηx =
dG ∂G y3
(0, 0, −1). The normal derivative of G is = − = with x = (x1 , x2 , 0), y =
dηx ∂x3 2πr3
(y1 , y2 , y3 ), y3 > 0, where r2 = (x1 − y1 )2 + (x2 − y2 )2 + y32 . Similarly we can check that
dG1
= 0, for x ∈ ∂Ω, y ∈ Ω.
dηx
The formula (3) suggests that a solution to the problem: ∆u = 0 in Ω, u = g on ∂Ω,
could be
Z ∞Z ∞
1 y3 g(x1 , x2 )dx1 dx2
Z
dG
u(y) = u(x) (x, y) dσx = ,
∂Ω dηx 2π −∞ −∞ [(x1 − y1 )2 + (x2 − y2 )2 + y32 ] 23

with x = (x1 , x2 , 0), g(x1 , x2 , 0) = g(x1 , x2 ), y = (y1 , y2 , y3 ), y3 > 0. (18)

1
Clearly G1 (x, y) = , for x = (x1 , x2 , 0), y ∈ Ω. Therefore, (3) suggests that a
2πkx − yk
du
possible solution to the problem ∆u = 0 in Ω, = g1 on ∂Ω, could be
dηx
Z ∞Z ∞
1 g1 (x1 , x2 )
Z
u1 (y) = − G1 (x, y)g1 (x)dηx = − dx1 dx2 , (19)
∂Ω 2π −∞ −∞ kx − yk
with x = (x1 , x2 , 0), g1 (x1 , x2 , 0) = g1 (x1 , x2 ), y = (y1 , y2 , y3 ), y3 > 0.
The functions g and g1 must guarantee the convergence of the improper integrals (18)
and (19), respectively (i.e., the existence of u and u1 ), and the fact that these functions u and
u1 are solutions of the above Dirichlet and Neumann problems with f = 0. An important
case in which these requirements are fulfilled is given by:

Theorem 3 Let g be continuous and bounded on R2 , and g1 be continuous and with compact
support in R2 . Then the functions u and u1 given by the improper integrals (18) and (19)
satisfy:
1. u ∈ C 2 (Ω) ∩ C(Ω̄)), ∆u = 0 in Ω, u = g on ∂Ω
du1
2. u1 ∈ C 2 (Ω) ∩ C 1 (Ω̄), ∆u1 = 0 in Ω, = g1 on ∂Ω.
dηx
Proof. We will use the formula
Z ∞Z ∞
1 y3 dx1 dx2
I0 = = 1, (20)
2π −∞ −∞ [(x1 − y1 ) + (x2 − y2 )2 + y32 ] 23
2
64 E. CONSTANTIN AND N. H. PAVEL

where x = (x1 , x2 , 0), y = (y1 , y2 , y3 ), y3 > 0.

To show (20), we make the change of variables x1 − y1 = y3 z, x2 − y2 = y3 w, so
dx1 = y3 dz, dx2 = y3 dw. The integral I0 becomes
Z ∞Z ∞
1 dzdw
I0 = .
2π −∞ −∞ (z + w2 + 1)3/2
2

Z
dt t
Since = √ + C, we get
(t2 + a2 )3/2 a t + a2
2 2

Z ∞ Z t2 
1 dz
I0 = lim lim dw
2π −∞ t1 →−∞ t2 →∞ t1 (z 2 + w2 + 1)3/2
Z ∞  
1 z t2
= lim lim √ | dw
2π −∞ t1 →−∞ t2 →∞ (w2 + 1) z 2 + w2 + 1 t1
Z ∞" #
1 t2 t1
= lim p − lim p dw
2π −∞ t2 →∞ (w2 + 1) t22 + w2 + 1 t1 →−∞ (w2 + 1) t21 + w2 + 1

1 ∞
Z t2
1 1 1
Z
= 2
dw = lim lim 2
dw
π −∞ w + 1 π 1
t →−∞ t 2 →∞ t1 w + 1

1
= lim lim (arctan t2 − arctan t1 ) = 1.
π t1 →−∞ t2 →∞

The fact that u and u1 are harmonic on Ω follows by differentiating under the integral
sign and taking into account that G1 (x, y) = G1 (y, x), x ∈ Ω̄, y ∈ Ω, x 6= y, ∆y G1 (x, y) =
∆y G1 (y, x) = 0, y ∈ Ω, x ∈ ∂Ω, and
dG d
∆y (x, y) = (∆y G(x, y)), G(x, y) = G(y, x) for y 6= x, x ∈ Ω̄, y ∈ Ω, ∆y G(x, y) =
dηx dηx
∆y G(y, x) = 0 for y ∈ Ω, x ∈ ∂Ω.
As g is bounded on R2 , and g1 is continuous and with compact support in R2 , u and u1
given by (18) and (19) are bounded due to Theorem 2.
Let us to prove that u(x) = g(x) for x = (x1 , x2 , 0), i.e., that

lim u(y) = g(ȳ), (21)

y→ȳ

for all ȳ = (ȳ1 , ȳ2 , 0), y = (y1 , y2 , y3 ), with y3 > 0.

In view of (18) and (20), we can write
Z ∞Z ∞
1 y3 (g(x1 , x2 ) − g(ȳ1 , ȳ2 ))dx1 dx2
u(y) − g(ȳ) = , (22)
2π −∞ −∞ [(x1 − y1 )2 + (x2 − y2 )2 + y32 ] 32

where y = (y1 , y2 , y3 ), y3 > 0, ȳ = (ȳ1 , ȳ2 , 0), x = (x1 , x2 , 0), (x1 , x2 ) ∈ R2 .

Given ǫ > 0, there exists δ(ǫ) > 0 such that
|g(x1 , x2 ) − g(ȳ1 , ȳ2 )| < ǫ, for k(x1 , x2 ) − (ȳ1 , ȳ2 )k < δ.
GREEN FUNCTION OF THE LAPLACIAN FOR THE NEUMANN PROBLEM 65

We obtain
1 y3 (g(x1 , x2 ) − g(ȳ1 , ȳ2 ))dx1 dx2
Z Z
u(y) − g(ȳ) = 3
2π k(x1 ,x2 )−(ȳ1 ,ȳ2 )k<δ [(x1 − y1 )2 + (x2 − y2 )2 + y32 ] 2
1 y3 (g(x1 , x2 ) − g(ȳ1 , ȳ2 ))dx1 dx2
Z Z
+ 3 = I1 + I2 . (23)
2π k(x1 ,x2 )−(ȳ1 ,ȳ2 )k≥δ [(x1 − y1 )2 + (x2 − y2 )2 + y32 ] 2
From (20), it follows that |I1 | ≤ ǫ.
We will show that the integral I2 converges to zero for y → ȳ.
δ
If ky − ȳk < , kx − ȳk ≥ δ, we get
2
δ 1
kx − ȳk ≤ kx − yk + ky − ȳk ≤ kx − yk + ≤ kx − yk + kx − ȳk,
2 2
1
and so kx − yk ≥ kx − ȳk.
2
Since g is bounded, let M > 0 be a constant such that |g(x)| ≤ M , for all x ∈ R2 .
Then
2M
Z Z
y3 dx1 dx2
|I2 | ≤ 2 3
2π 2 2
k(x1 ,x2 )−(ȳ1 ,ȳ2 )k≥δ [(x1 − y1 ) + (x2 − y2 ) + y3 ] 2

8M y3
Z Z
dx1 dx2
≤ → 0,
π 2 2 3
k(x1 ,x2 )−(ȳ1 ,ȳ2 )k≥δ [(x1 − ȳ1 ) + (x2 − ȳ2 ) ] 2
as y → ȳ, i.e., as y3 → 0+, because the above integral is bounded in view of Theorem 2.
We conclude that lim I2 = 0, which in conjunction with |I1 | ≤ ǫ implies lim sup |u(y) −
y→ȳ y→ȳ
g(ȳ)| ≤ ǫ, for all ǫ > 0, so (21) is valid.
Under the hypotheses on g1 the function u1 is well defined and of class C 1 (Ω̄). Differen-
tiating under the integral sign we obviously obtain:
Z ∞Z ∞
∂u1 (y) 1 y3 g1 (x1 , x2 )dx1 dx2
− = .
∂y3 2π −∞ −∞ [(x1 − y1 )2 + (x2 − y2 )2 + y32 ] 23
On the basis of the previous discussion, this implies:
du1 ∂u1 (y)
=− = g1 (y) on the boundary ∂Ω of Ω as g1 is bounded, which completes the
dηy ∂y3
proof.

3.3 Construction of the Green function for the Neumann Problem

for Rn+
Next we generalize the result of section 3.2 by constructing the Green function for the Neu-
mann problem for the positive half-space Rn+ , n > 2.

Let Ω = Rn+ = {x = (x1 , x2 , . . . , xn ) ∈ Rn , xn > 0}, n > 2, be the positive half-space.

Clearly, ∂Ω = {x = (x1 , x2 , . . . , xn ) ∈ Rn , xn = 0} = {x = (x1 , . . . , xn−1 , 0), (x1 , . . . , xn−1 ) ∈
Rn−1 }.
For y = (y1 , y2 , . . . , yn ) define by reflection y ∗ = (y1 , . . . , yn−1 , −yn ).
Then the function:
66 E. CONSTANTIN AND N. H. PAVEL

G(x, y) = −ψ(x, y) + ψ(x, y ⋆ ), x ∈ Ω̄, y ∈ Ω, x 6= y, i.e.,

1 1
G(x, y) = n−2
− , (24)
(n − 2)σn kx − yk (n − 2)σn kx − y ∗ kn−2

is a Green function for the (DP) with f = 0, and

G1 (x, y) = −ψ(x, y) − ψ(x, y ⋆ ), x ∈ Ω̄, y ∈ Ω, x 6= y, i.e.,

1 1
G1 (x, y) = n−2
+ , (25)
(n − 2)σn kx − yk (n − 2)σn kx − y ∗ kn−2

is a Green type function for (NP) with f = 0.

As kx − yk = kx − y ∗ k = r for y ∈ Ω, x ∈ ∂Ω, it follows that G(x, y) = 0, x ∈ ∂Ω, y ∈ Ω.
The outward normal ηx to ∂Ω at x ∈ ∂Ω is ηx = (0, . . . , 0, −1). The normal derivative
dG ∂G 2yn
of G is =− = , x = (x1 , . . . , xn−1 , 0), y = (y1 , . . . , yn−1 , yn ), yn > 0, where
dηx ∂xn σn r n
dG1
r2 = (x1 − y1 )2 + . . . + (xn−1 − yn−1 )2 + yn2 . Similarly we can check that = 0, for
dηx
x ∈ ∂Ω, y ∈ Ω.
The formula (3) suggests that a solution to the problem: ∆u = 0 in Ω, u = g on ∂Ω,
could be:

2yn
Z Z
dG g(x)
u(y) = u(x) (x, y) dσx = dσx
∂Ω dη x σ n n
∂R+ kx − ykn

∞ ∞
2yn g(x1 , . . . , xn−1 )dx1 . . . dxn−1
Z Z
= ... n , (26)
σn −∞ −∞ [(x1 − y1 )2 + . . . + (xn−1 − yn−1 )2 + yn2 ] 2

x = (x1 , . . . , xn−1 , 0), g(x1 , . . . , xn−1 , 0) = g(x1 , . . . , xn−1 ), y = (y1 , . . . , yn ), yn > 0.

The function u is the solution of the (DP) with f = 0 (see [4] for the proof of the Poisson’s
formula for the half-space of Rn+ , n > 2).
2
Clearly G1 (x, y) = , for x = (x1 , . . . , xn−1 , 0), y ∈ Rn+ .
(n − 2)σn kx − ykn−2
du
Therefore, (3) suggests that a possible solution to the problem ∆u = 0 in Ω, = g1
dηx
on ∂Ω, could be

2 g1 (x)
Z Z
u1 (y) = − G1 (x, y)g1 (x)dσx = − dσx , (27)
+
∂Rn (n − 2)σn ∂R+ kx − ykn−2
n

x = (x1 , . . . , xn−1 , 0), g1 (x1 , . . . , xn−1 , 0) = g1 (x1 , . . . , xn−1 ), y = (y1 , . . . , yn ), yn > 0.

The function g1 must guarantee the convergence of the improper integral (27) (i.e., the
existence of u1 ), and the fact that this function u1 is solution of the Neumann problems
with f = 0. An important case in which these requirements are fulfilled is given by:
GREEN FUNCTION OF THE LAPLACIAN FOR THE NEUMANN PROBLEM 67

Theorem 4 Let g1 be continuous and with compact support in Rn−1 .

Then the function u1 given by the improper integral (27) satisfies:
du1
u1 ∈ C 2 (Ω) ∩ C 1 (Ω̄), ∆u1 = 0 in Ω, = g1 on ∂Ω.
dηx
Proof. We will use the formula
2yn 1
Z
I0 = dσx = 1, (28)
σn ∂Rn
+
kx − ykn

where x = (x1 , . . . , xn−1 , 0), y = (y1 , . . . , yn ), yn > 0.

The fact that u1 are harmonic on Ω follows by differentiating under the integral sign
and taking into account that G1 (x, y) = G1 (y, x), x ∈ Ω̄, y ∈ Ω, x 6= y, ∆y G1 (x, y) =
∆y G1 (y, x) = 0, y ∈ Ω, x ∈ ∂Ω.
As g1 is continuous and with compact support in Rn−1 , u1 defined by (27) is bounded
due to Theorem 2.
du
Let us to prove that = g1 (x) for x = (x1 , . . . , xn−1 , 0), i.e., that
dηx
du
lim = g1 (ȳ), (29)
y→ȳ dηy

for all ȳ = (ȳ1 , . . . , ȳn−1 , 0) fixed, and y = (y1 , . . . , yn ), yn > 0.

Under the hypotheses on g1 the function u1 is well defined and of class C 1 (Ω̄). Differen-
tiating under the integral sign we obviously obtain:

∂u1 (y) 2 yn g1 (x)

Z
du1
=− = dσx .
dηy ∂yn σn ∂Rn+ kx − ykn

In view of (27) and (28), we can write

2yn g1 (x) − g1 (ȳ)

Z
du1
− g(ȳ) = dσx ,
dηy σn ∂Rn
+
kx − ykn

where y = (y1 , . . . , yn ), yn > 0, ȳ = (ȳ1 , . . . , ȳn−1 , 0), x = (x1 , . . . , xn−1 , 0), (x1 , . . . , xn−1 ) ∈
Rn−1 .
Given ǫ > 0, there exists δ(ǫ) > 0 such that
|g1 (x) − g1 (ȳ)| < ǫ, for kx − ȳk < δ.
δ
Then if ky − ȳk < , y ∈ Rn+ , we have
2
2yn g1 (x) − g1 (ȳ) 2yn g1 (x) − g1 (ȳ)
Z Z
du1
− g(ȳ) = n
dσ x + dσx
dηy σn ∂Rn+ ∩B(ȳ,δ) kx − yk σn ∂Rn+ \B(ȳ,δ) kx − ykn

= I1 + I2 , (30)
68 E. CONSTANTIN AND N. H. PAVEL

where B(ȳ, δ) = {x ∈ Rn ; kx − ȳk < δ}.

From (28), it follows that |I1 | ≤ ǫ.
δ
If ky − ȳk < , kx − ȳk ≥ δ, we get
2
δ 1
kx − ȳk ≤ kx − yk + ky − ȳk ≤ kx − yk + ≤ kx − yk + kx − ȳk,
2 2
1
and so kx − yk ≥ kx − ȳk.
2
Since g1 is bounded, let M > 0 be a constant such that |g1 (x)| ≤ M , for all x ∈ Rn−1 .
Then
2yn
Z
|I2 | ≤ 2M dσx
∂Rn
+ \B(ȳ,δ)
σn kx − ykn

2n+2 M yn 1
Z
≤ dσx → 0,
σn n
∂R+ \B(ȳ,δ) kx − ȳkn
as y → ȳ, i.e., as yn → 0+, because the above integral is bounded in view of Theorem 2.
du
We conclude that lim I2 = 0, which in conjunction with |I1 | ≤ ǫ implies lim sup | −
y→ȳ y→ȳ dηy
g1 (ȳ)| ≤ ǫ, for all ǫ > 0, so (29) is valid, which completes the proof.

References
[1] Barbu, Viorel, Partial differential equations and boundary value problems, Kluwer Academic Publish-
ers, Dordrecht ; Boston, Mathematics and its applications ; Vol. 441, 1998.

[2] Ciric, I.R., Green functions of the Laplacian for Neumann problems in circular cylindrical coordinates,
Collection of articles dedicated to Remus Radulet on his 70th Birthday, I. Rev. Roumaine Sci. Tech.
Ser. Electrotech. Energt. 19, 1974, 547-560.

[3] Dibenedetto, E., Partial Differential Equations, Second Edition, Birkhauser, Boston, 2010.

[4] Evans, L.C., Partial Differential Equations, AMS, Graduate Studies in Mathematics, Vol. 19, 1998.

[5] Hazlett, R.D.; Babu, D.K., Readily computable Green’s and Neumann functions for symmetry-
preserving triangles, Quarterly of Applied Mathematics, 67, no. 3, 2009, 579–592.

[6] Hitotumatu, Sin, On the Neumann function of a sphere, Comment. Math. Univ. St. Paul. 3, 1954, 1-5.

[7] Ivanov, V.I., and Trubetskov, M.K., Handbook of Conformal Mapping with Computer-Aided Visual-
ization, CRC Press, Inc., Boca Raton, Florida, 1995.

[8] John, Fritz, Partial Differential Equations, Springer-Verlag, Applied Mathematical Sciences, Vol.1, 4th
Edition, 1991.

[9] Kevorkian, J., Partial Differential Equations: Analytical Solution Techniques, Wadsworth, Inc., Bel-
mont, California, 1990.

[10] Lin, C.C., and Segal, L.A, Mathematics Applied to Deterministic Problems in the Natural Sciences,
SIAM, 1988.

[11] McCann, R. C., Hazlett, R.D., and Babu, D.K., Highly accurate approximations of Green’s and Neu-
mann functions on rectangular domains, R. Soc. Lond. Proc. Ser. A Math. Phys. Eng. Sci. 457, no.
2008, 2001, 767–772.
GREEN FUNCTION OF THE LAPLACIAN FOR THE NEUMANN PROBLEM 69

[12] Nayar, B.M., Neumann function for the sphere, I, II, Indian J. Pure Appl. Math. 12, no. 10, 1981,
1266–1282, 1283–1292.

[13] Newman, J.N., The Green function for potential flow in a rectangular channel, Journal of Engineering
Mathematics, 26, 1992, 51-59.

[14] Pavel, Nicolae H., Introduction to Partial Differential Equations, Zip Publishing, Columbus, Ohio,
2007. www.zippublishing.com.

[15] Pinchover, Y., and Rubinstein, J., An Introduction to Partial Differential Equations, Cambridge Uni-
versity Press, Cambridge, UK, 2005.

[16] Rauch, Jeffrey, Partial Differential Equations, Springer-Verlag, Graduate Texts in Mathematics, Vol.
128, 1991.

[17] Roach, Gary F., Green’s Functions: Introductory Theory with Application, Van Nostrand Reinhold
Company, London, 1970.

[18] Taylor, M.E., Partial Differential Equations, Springer-Verlag, Applied Mathematical Sciences, Vol.
115, 1996.

[19] Sobolev, S.L., Partial Differential Equations of Mathematical Physics, New York, Dover Publications,
Inc., 1994.